So what these results tell us is that a large MOV in home games predicts a large MOV in road games? Hm. It is amazingly linear but I don't find much surprising about it, at all.
Nor should you, yet the people who set up brackets and put teams in championship games insist upon ignoring this fact to the largest degree possible. Note that the point of KenPom's study was NOT about winning at home and on the road . . . it was one win predicting another with as much held constant as possible-- in this case playing the same team.
As for your comments on the BCS, you neglected to mention that the logic of removing MOV from the computer polls is that the human polls do factor it in, and those out-weigh the computers anyway.
To my recollection that was not-- at all-- part of the reasoning for removing it. The fact that it was very important to voters was actually ignored in the process. It was an attempt at influencing the behavior of coaches in the case of mismatches on the football field . . . to remove the incentive to RUTS. And it didn't work at all since, as you noted, the voters love RUTS more than computers ever could. This was NOT an effort to improve the accuracy of the rankings; rather it was an effort to not encourage hurt feelings. And it was an ignorant change no matter how you look at it.
I suspect football is less correlated, due to what I would call the "non-linear scoring". That is, one TD is much more of a difference-maker than one bucket. Also, football games probably tend to get out of hand earlier, leading to blowouts that might be worse than they look. Of course teams score in garbage time also even when down big, so maybe that part of it evens out. But I think the larger "basic unit" score in football would lead to a less linear result, and my subjective experience from the past seems to confirm this. It's not unusual at all to see a close game between two good teams but a blowout between one of the same teams and a different (on paper) equivalent team.
I don't think that would make the association weaker (though there's no guarantee it would be linear); it would just add to the variance. If anything I would think the association would be more consistent since there is less parity in football. It's hard to say and pretty much impossible to demonstrate with data, though.
And of course rematches are very rare in college football, but LSU-bammer from last year's BCS was definitely not a predictor.
No, but I believe football is far more deterministic than basketball. Only in very closely matched teams in football would I say the loser of the first game would have more or less even chances of winning the second game. In basketball the loser of the first game very, very frequently wins the second and not necessarily because the teams are even. The effect of one or two players on a basketball game cannot be overstated and the result is a much wider variety of upset victories. So in that sense, the tight clustering around the line in the above graphs should be more surprising than our initial reaction, and for the same reason we should see an even tighter relationship in football (sans the variance due to scoring in larger clumps) since there are fewer upsets between disparate teams.
But of course this is speculation.
